Like Gold Rush sluice boxes that separate gold from fool’s gold, statistics lets real discoveries glitter instead of masquerading background events.

Statistics played an important role in CERN’s decision three years ago to turn off the LEP machine and begin on-time construction of a new machine. Even though there were tantalizing hints of the still unearthed Higgs particle detected just before LEP was to turn off, "statistically, there weren’t enough events and the confidence limits were not as strong as needed to claim a discovery," said Louis Lyons, a particle physicist at Oxford. CERN decided not to make the expensive changes to its construction contracts that would be required to keep the machine on.

In order to add advanced statistics techniques to the toolbox of particle physicists, astrophysicists and cosmologists, some of the world’s greatest statisticians came to SLAC from September 8 to 11 for the PHYSTAT 2003 conference.

About 120 participants studied advanced statistics for measurements and searches in their fields, hoping to improve results and save time and frustration when analyzing the mounds of data accumulated from an experiment.

"We spend a lot of time, effort and money to build, design and run apparatus. Getting the most out of your data is very important and relatively cheap," said conference organizer Lyons. He is comfortable wielding statistical tools, and has written a book and given many lectures on the topic.

"Even particle physicists can find statistics a chore," he said, "but it’s an essential part of correctly understanding what an experiment has measured and to what accuracy." Statistics is also used to set a limit on probabilities, check if the data is consistent with Standard Model predictions, and combine results from different experiments to create a more sensitive answer.

"We learn statistics the hard way, by trying it out. The conference really was meant to enhance the statistical ability of people in these fields," said Lyons. "In particle physics, astrophysics and cosmology, people work with different tools—accelerators versus telescopes—but nonetheless a lot of the data analysis techniques are very similar."

Compared to previous physics conferences on statistics, the SLAC event broadened its audience to the astrophysics and cosmology community, and invited more statisticians to provide expert insight.

Stanford Statistics Professor Brad Efron gave the keynote address, entitled "Bayesians, Frequentists and Physicists," about different approaches to statistics. Efron is also president of the American Statistical Association and a MacArthur Prize winner.

Seth Digel (GLAST) and Frank Porter (BaBar) gave talks on statistical issues they face. Jerry Friedman (SCS), a particle physicist turned professor of statistics at Stanford, spoke on "Modern Developments in Machine Learning."

Setting limits is an important statistical tool. Many experiments look for things but don’t see them, like the search for dark matter and the Higgs particle.

"But rather than say you don’t see it, you can say the maximum effect that could be there is x," said Lyons. It’s similar to learning an item is smaller than a breadbox when playing 20 Questions. "If you set a good limit, it can be very significant."

For example, though LEP did not find the Higgs particle, its search was very sensitive. Physicists can now say the Higgs particle, if it exists, has a mass heavier than 114 GeV. LEP’s successor (LHC), Fermilab’s Tevatron, and any future Linear Collider will have shots at finding the particle at those greater energies.

Other conference topics included signal significance, systematics, spatial data, non-parametric estimation, unfolding convolution, blind analyses, multivariate classification, variability of sources, hypothesis testing, goodness of fit and cluster analysis.

The local conference committee consisted of Richard Mount, Arla LeCount, Joseph Perl and David Lee. Local members of the scientific committee were Roger Barlow, Seth Digel, Brad Efron, Jerry Friedman, Jeffrey Scargle and Steve Yellin.

For more information, and a recommended reading list, see: http://www-conf.slac.stanford.edu/phystat2003/